This project studies the production of the human voice and the oscillatory regimes resulting from the laryngeal and vocal tract articulation (involved in, e.g., modal or falsetto voice registers). It relies on physical modeling and nonlinear systems theory to elaborate simulations, observers-controllers and bifurcations analyses of vocal apparatus. Its first specificity is to develop these tools under the systematic hypothesis of guaranteed power balances (preventing non-physical instabilities). The second is to complete, feed and assess them by measurements both in vivo and on a coherent set of mechatronic testbeds (with control of ex vivo, biomimetic or standardized larynx, and of artificial vocal tracts). The objective is to scientifically understand, reproduce and analyze the voice (in healthy or pathological cases) using virtual and robotic avatars, with possible artistic and medical spin-offs.